Abstract: A new support vector multiclassification methodology was proposed where several binary support vector binary classifiers, each of which equipped with a feature extractor based on kernel principle components analysis (Kernel PCA), were organized in a serial structure. Its training and classification algorithm was described. The BP net classifier, RBF net classifier, traditional support vector multiclassifier and serial support vector multi-classifier (SSVC) were used for analog circuit fault diagnosis. Compared with BP net and RBF net classifiers, support vector approach led to significantly better classification accuracy on test patterns. The SSVC afforded top diagnosis accuracy among these classifiers and outperforms traditional support vector multicalssifier dramatically in training and classification efficiency.